On Sat, Sep 08, 2012 at 10:01:13AM +0200, Ralf Hildebrandt wrote:
> When sending lots of mails (mass mailings) via many machines, one
> quickly realizes that the current concept of smtp_fallback_relay is
> a bit problematic:
If one thinks harder, one realizes the purpose of the mechanism is
to move messages (to destinations) that gum up the queue out of
the primary queue, where they may impact the latency of delivery
to inocent destinations.
Nobody said that the fallback relay has to another machine. You
can and should in many cases configure a second Postfix instance
on each machine to be the fallback relay, this solves the greylisting
by IP problem, and keeps the fallback load distributed to all the
available hardware.
In the fallback queue (instance) increase the active queue limits,
and delivery agent process limits, since you expect this mail to
generate less network traffic per delivery attempt and to incur
a larger active queue size due to the longer queue occupancy per
message.
There is no need to redesign Postfix, in fact getting the junk out
of the primary queue is always preferable.
--
Viktor.
